Many people assume AI is expensive, or that the pricing is so complicated it’s better to wait. In reality there are two ways to pay, and most small businesses only need one of them.

1. Subscription: a fixed price per person

For everyday work, like writing emails, reviewing contracts, summarising meetings or working in spreadsheets, you pay a fixed monthly subscription. Claude’s prices (in US dollars, excluding tax):

  • Free: no cost, with limited usage. Good for trying it out.
  • Pro: $20 a month, or $17 if you pay for the year. Enough for most sole traders.
  • Max: from $100 a month, for people who use AI all day.
  • Team: for 2 to 150 people, from $20 per seat per month billed annually. Central admin and billing, and your data isn’t used for training by default.

Rule of thumb: start with Pro for one or two people, and move to Team when more join and you want to manage access in one place.

2. Pay as you go: when automation takes over

If AI is going to work automatically, for example sorting every incoming invoice or answering enquiries from your website, you pay for usage through the API. The price is counted in tokens, which are pieces of words. A million tokens is roughly 750,000 words of English.

The price depends heavily on the model. Claude Haiku 5.5, released on 7 October, costs $0.10 per million input tokens and $0.50 per million output tokens. Opus 5.5 costs $4 in and $20 out. For simple, repetitive work like classification that makes a big difference: sorting a thousand short invoices with Haiku usually costs less than a dollar.

Two tips bring the bill down further:

  • Batch: work that can wait a few hours gets a 50% discount.
  • Prompt caching: if the same instructions are sent again and again, re-reading them costs only a fraction of the full price.

3. The real cost is time

The subscription is rarely the biggest item. The bigger cost is the time it takes to learn to use the tool well, or the time lost if nobody does. One hour a week saved on bookkeeping or payroll pays for a Pro subscription many times over.

So start with one task you do every week, measure the time before and after, and then decide whether it’s worth expanding.

References